sched_xetaffinity test broken on compat

[Dmitry V. Levin]

On Tue, Aug 29, 2023 at 01:33:10PM +0300, Dmitry V. Levin wrote:
> On Mon, Jul 31, 2023 at 09:39:51AM +0200, Sven Schnelle wrote:
> > Hi List,
> > 
> > 17c0945526ba ("tests/sched_xetaffinity: add a check for a
> > non-properly-sized CPU set") seems to be broken on compat.
> > I see the following both on s390 and x86:
> > 
> > 64 bit:
> > 
> > $ ./sched_xetaffinity
> > sched_getaffinity(167052, 1, NULL) = -1 EINVAL (Invalid argument)
> > sched_getaffinity(167052, 2, NULL) = -1 EINVAL (Invalid argument)
> > sched_getaffinity(167052, 4, NULL) = -1 EINVAL (Invalid argument)
> > sched_getaffinity(167052, 8, NULL) = -1 EFAULT (Bad address)
> > sched_getaffinity(167052, 8, 0x7f5ad2b4a078) = -1 EFAULT (Bad address)
> > sched_getaffinity(167052, 8, [0 1 2 3 4 5 6 7 8 9 10 11]) = 8
> > sched_setaffinity(167052, 0, []) = -1 EINVAL (Invalid argument)
> > sched_setaffinity(167052, 1, 0x7f5ad2b4a000) = -1 EFAULT (Bad address)
> > sched_setaffinity(167052, 7, [8]) = 0
> > sched_setaffinity(167052, 8, [11]) = 0
> > sched_getaffinity(167052, 128, [11]) = 8
> > +++ exited with 0 +++
> > 
> > Which looks ok, but on compat:
> > 
> > $ ./sched_xetaffinity
> > sched_getaffinity(166994, 1, NULL) = -1 EINVAL (Invalid argument)
> > sched_getaffinity(166994, 2, NULL) = -1 EINVAL (Invalid argument)
> > sched_getaffinity(166994, 4, NULL) = -1 EFAULT (Bad address)
> > sched_getaffinity(166994, 4, 0xf7ee907c) = -1 EFAULT (Bad address)
> > sched_getaffinity(166994, 4, [0 1 2 3 4 5 6 7 8 9 10 11]) = 4
> > sched_setaffinity(166994, 0, []) = -1 EINVAL (Invalid argument)
> > sched_setaffinity(166994, 1, 0xf7ee9000) = -1 EFAULT (Bad address)
> > sched_setaffinity(): Bad address
> > 
> > I think this is because the test does:
> > 
> >                 if (setaffinity(pid, 7, cpuset))
> >                        perror_msg_and_skip("sched_setaffinity()");
> > 
> > which works on 64 bit, but not with 32 bit. Instead of returning skip(77)
> > i think the test should either be fixed, or this part should be skipped
> > on compat. Any thoughts?
> 
> I agree it would be nice to have it fixed, but I'm not quite familiar with
> the subject so I'm not sure what would be the proper fix.

Looks like it depends on the kernel, for me it works everywhere except
gcc14.fsffrance.org where it fails the way you described.


-- 
ldv